ESSAYS ON MIGRATION
by
ERKAN DUMAN
Submitted to the Institute of Social Sciences in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
Sabancı University
July 2018
© Erkan Duman 2018
All Rights Reserved
iv ABSTRACT
ESSAYS ON MIGRATION
ERKAN DUMAN Ph.D., Dissertation, July 2018
Dissertation Supervisor: Prof. Abdurrahman B. Aydemir
Keywords: remittances; child human capital accumulation; adult labor supply; migrant networks; location choice
This dissertation includes two chapters. In the first chapter, we examine the impacts of remittances on various household outcomes including: child school attendance and child illiteracy, child labor, adult labor and household well-being. We use IV estimation technique to account for the endogeneity of remittances. We find evidence of a significant positive impact of remittances on school attendance of 6- to 19-years-old boys and of 6- to 14-years-old girls.
Receiving remittances leads to a lower school retention for 15- to 19-years-old girls. Girls of ages 6-to-14 from recipient households are more likely to be literate. Children aged 15 to 19 in recipient households are significantly less likely to supply labor. Adult labor supply results are in favor of income effect hypothesis. Lastly, recipient households are shown to be relatively better-off with respect to welfare compared to non-recipients. In the second chapter, we estimate the determinants of 28-54-years-old male work migrants’ location choices among 67 provinces of Turkey. The results show that internal migrants respond to differences in migrant networks, labor market and population attributes between locations while deciding on the migration destination. Distance from the source province is shown to be a significant deterrent of immigrant’s location choice. Migrants are drawn to cities in which their former compatriots are highly concentrated. Migrants are more likely to move to cities which have relatively better economic conditions as captured by lower unemployment rate and to cities with larger populations that has larger economic activity.
v ÖZET
GÖÇ ÜZERİNE MAKALELER
ERKAN DUMAN Doktora Tezi, Temmuz 2018
Tez Danışmanı: Prof. Dr. Abdurrahman B. Aydemir
Anahtar Kelimeler: uluslararası para transferleri; çocuk beşeri sermaye birikimi; yetişkin emek arzı; göçmen ağları; yer seçimi
Bu tez iki bölüm içermektedir. İlk bölümde, uluslararası para transferlerinin hanehalklarındaki çocukların okul devamsızlıkları, okur-yazar olma durumları, ve emek arzları; hanehalklarındaki yetişkinlerin emek arzları; hanehalklarının refah düzeyleri üzerine etkisi araştırılmaktadır. Para transferlerinin içselliğini hesaba katmak için araç değişken yöntemi kullanılmaktadır. 6-19 yaş grubu erkekler ve 6-14 yaş grubu kızların okul devam durumları üzerinde para transferlerinin pozitif manidar etkisine delil bulunmuştur. Para transferi almak 15-19 yaş grubu kızların okul devam ihtimallerini azaltmaktadır. Para transferi alan hanelerdeki 6-14 yaş grubu kızların, para transferi almayan hanelerdeki emsallerine kıyasla okur-yazar olma ihtimalleri daha yüksektir.
Para transferi alan hanelerdeki 15-19 yaş grubu çocukların emek arz etme ihtimalleri daha düşüktür. Yetişkin emek arzı sonuçları gelir etkisi hipotezinin ağır bastığına delalet etmektedir.
Son olarak, para transferi alan hanelerin almayan hanelere kıyasla refah düzeyi bakımından daha iyi durumda olduğu görülmüştür. İkinci bölümde, 28-54 yaş grubu iş amaçlı göç eden erkeklerin Türkiye’deki 67 il arasından yer seçimlerinin belirleyicilerini tahmin etmeye çalışıyoruz. Sonuçlar göstermektedir ki iç göçmenler yer seçimlerini yaparken illerin göçmen ağları, emek piyasaları ve popülasyon özellikleri farklarını göz önünde bulundurmaktadır.
Kaynak ilden mesafenin iç göçmenlerin yer seçimleri önünde manidar bir engelleyici olduğu
bulunmuştur. İç göçmenler hemşehrilerinin yoğun yaşadığı illere çekilmektedirler. İç
göçmenler ekonomik bakımdan daha iyi konumda olan illere ve popülasyonları daha büyük
olan illere göç etmeyi tercih etmektedirler.
vi
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to my thesis supervisor Prof. Dr.
Abdurrahman B. Aydemir for his guidance in all phases of the thesis and his help during the course of my Ph.D. studies. His support was priceless for me. Without his support, it would have been impossible for me to finish my Ph.D.
I would also like to thank Prof. Dr. Murat G. Kırdar for his insightful comments and suggestions.
I am deeply grateful to my wife Hatice Duman for her moral support and endless
encouragement.
vii
TABLE OF CONTENTS
1 THE IMPACTS OF REMITTANCES ON CHILD SCHOOL ATTENDANCE AND
ILLITERACY, CHILD LABOR, ADULT LABOR AND HOUSEHOLD WELL-BEING:
EVIDENCE FROM TURKEY ... 1
1.1 Introduction ... 1
1.1.1 Migration and remittance history of Turkey ... 8
1.2 Identification Strategy and Estimation Methodologies ... 10
1.2.1 Econometric Identification ... 10
1.2.2 Other Estimation Issues ... 20
1.3 Data and Summary Statistics ... 33
1.3.1 Data and Sample Definition ... 33
1.3.2 Descriptive Statistics ... 37
1.4 Results ... 41
1.4.1 Determinants of remittances ... 42
1.4.2 Main Results ... 49
1.4.2.1 Child human capital investment decisions ... 49
1.4.2.1.1 Child school attendance and illiteracy ... 52
1.4.2.1.2 Child labor ... 59
1.4.2.2 Adult labor supply responses ... 63
1.4.2.2.1 Labor supply responses of adult males ... 63
1.4.2.2.1.1 20-24-years-old males ... 63
1.4.2.2.1.2 25-49-years-old males ... 65
1.4.2.2.1.3 50-64-years-old males ... 67
1.4.2.2.2 Labor supply responses of adult females ... 69
1.4.2.2.2.1 20-24-years-old females ... 70
1.4.2.2.2.2 25-49-years-old females ... 72
1.4.2.2.2.3 50-64-years-old females ... 74
1.4.2.3 Remittances and the welfare of households ... 75
1.5 Conclusion ... 80
Appendix ... 134
2 THE IMPACT OF MIGRANT NETWORKS ON IMMIGRANTS’ LOCATION CHOICES .. 136
2.1 Introduction ... 136
2.2 Methodology ... 143
2.3 Data and Descriptive Statistics ... 158
2.4 Results ... 162
2.5 Robustness Checks ... 168
2.6 Conclusion ... 174
REFERENCES ... 190
viii
LIST OF TABLES
Table 1-1 Distribution of remittance receipts and amount (average per year) ... 83
Table 1-2 Descriptive statistics of key variables for households with a child aged 6 to 19 ... 84
Table 1-3 Descriptive Statistics of key variables for households with an adult aged 20 to 64 ... 86
Table 1-4 Descriptive statistics of key variables for households ... 88
Table 1-5 First stage estimations (child sample) ... 90
Table 1-6 First stage estimations (samples of working age adult males) ... 93
Table 1-7 First stage estimations (samples of working age adult females) ... 95
Table 1-8 First stage estimations (sample of households) ... 97
Table 1-9 The impact of remittances on school attendance of children aged 6 to 14 ... 99
Table 1-10 The impact of remittances on child illiteracy (ages 6-14 years old) ... 101
Table 1-11 The impact of remittances on school attendance of children aged 15 to 19 ... 103
Table 1-12 The impact of remittances on child labor (boys aged 15 to 19) ... 105
Table 1-13 The impact of remittances on child labor (girls aged 15 to 19) ... 107
Table 1-14 The impact of remittances on child labor (girls aged 15 to 19 – models omit controls for
regional labor market characteristics) ... 109
Table 1-15 The impact of remittances on adult labor (males aged 20 to 24) ... 111
Table 1-16 The impact of remittances on adult labor (females aged 20 to 24) ... 113
Table 1-17 The impact of remittances on adult labor (males of ages 20-24 years old who currently live
with their parents) ... 115
Table 1-18 The impact of remittances on adult labor (females of ages 20-24 years old who currently
live with their parents) ... 117
Table 1-19 The impact of remittances on adult labor (males aged 25 to 49) ... 119
Table 1-20 The impact of remittances on adult labor (females aged 25 to 49) ... 121
Table 1-21 The impact of remittances on adult labor (males aged 50 to 64) ... 123
Table 1-22 The impact of remittances on adult labor (females aged 50 to 64) ... 125
Table 1-23 The impact of remittances on household well-being – part 1... 127
Table 1-24 The impact of remittances on household well-being – part 2... 129
Table 1-25 Treatment effects of receiving remittances on outcomes ... 131
Table A1 Reduced form regressions for non-receiving samples ... 134
Table 2-1 Descriptive Statistics ... 177
Table 2-2 Determinants of location choice - using 28-54 years old male work migrants ... 178
Table 2-3 Determinants of location choice – omitting province-group dummies ... 179
Table 2-4 Determinants of location choice – alternative measure for labor market condition - 1... 180
Table 2-5 Determinants of location choice – alternative measure for labor market condition - 2... 181
Table 2-6 Determinants of location choice – population density as alternative population control ... 182
Table 2-7 Determinants of location choice – land area added to the main specification... 183
Table 2-8 Determinants of location choice – foreign-born share of province omitted ... 184
Table 2-9 Determinants of location choice – networks based on living in the same origin province - 1
... 185
Table 2-10 Determinants of location choice – networks based on living in the same origin province - 2
... 186
Table 2-11 Determinants of location choice – using only destination characteristics as regressors .. 187 Table 2-12 Determinants of location choice – including non-migrants to the estimation sample - 1 188 Table 2-13 Determinants of location choice – including non-migrants to the estimation sample - 2 189
1
1 THE IMPACTS OF REMITTANCES ON CHILD SCHOOL ATTENDANCE AND ILLITERACY, CHILD LABOR, ADULT LABOR AND HOUSEHOLD
WELL-BEING: EVIDENCE FROM TURKEY
1.1 Introduction
With the increase in international migration all over the world, an economic actor paves its way to the stage as an important international financial flow to developing countries-namely, remittances. The beginning of the 1990s witnessed remittances gaining importance over other international financial flows (e.g., foreign direct investment, portfolio investment, and official development assistance) to developing countries. Since the late 1990s, international migrants’
remittances have surpassed official development assistance and portfolio investment, and in the beginning of the 2000s, remittances have come very close to the total amount of foreign direct investment flows (Yang, 2011). In 2004, the estimated value of workers’ remittances to developing countries was $160 billion, with $40 billion going to Latin America (Acosta, 2006).
In 2009 and 2010, remittances to developing countries were $325 billion and $307 billion in nominal terms, respectively (Yang, 2011). The average annual real growth rate of remittances in the period 1999-2008; the decade preceding the 2008 financial crisis, is worthwhile mentioning: while foreign direct investment and official development assistance had average annual real growth rates of 11.0 percent and 5.8 percent respectively in the corresponding period, remittances exceeded both with an average annual real growth rate of 12.9 percent (Yang, 2011).
The excessive amounts of remittances sent to developing countries in the preceding
decades and its continued growth has attracted attention of researchers. Motivations behind the
decision to remit and development impact of remittances constitute the two broad areas on
remittances in the literature. Studies focusing on the former one suggest a number of motives
including altruism, exchange for the services provided to the migrant by recipients, insurance,
loan repayment, and investment (Brown and Poirine, 1997; Docquier and Rapoport, 2006). Pure
self-interest in the form of aspiration to receive inheritance can be added to the list as an
important goal in remitting especially when the remittances are sent to the parents’ of the
migrant and the inheritance is conditional on the behavior of the children (Lucas and Stark,
1985).
2
Another set of papers study the uses of remittances and simply ask how remittances affect recipient countries or households. Studies trying to find causal linkages between remittances and economic performance at the country level are inconclusive. Faini (2007) finds a positive relationship between remittances and economic growth; however, others find no or a negative relationship (Chami, Fullenkamp, and Jajah, 2003; Giuliano and Ruiz-Arranz, 2005).
Studies using micro level data are partly motivated by the desire to understand remittance impacts in greater detail. In studies on household level impacts of remittances, choices made by the households with respect to the usage of remittances on consumption and/or investment expenditures are frequently observed. There is no widely accepted view on which of these two-alternative use of remittances is desirable. Yang (2011) states that it could be optimal to use remittances on consumption where households suffer from low income levels;
whereas, it could be optimal to use remittances on productive investments for households that enjoy a sufficient or a higher wealth level and where productive investments would not have been achieved due to budget constraints without the extra income derived from remittances.
Brown and Ahlburg (1999) conclude that increased income derived from remittances is used for higher levels of consumption in South Pacific island states. Yang (2008) shows that there is no correlation between an increase in remittances due to international migrants’ favorable exchange rate shocks and consumption expenditures of migrants’ origin households in Philippines. However, the exogenous increase in income leads to increased entry into capital intensive enterprises such as transportation and manufacturing by the migrants’ origin households in this context.
Investing in the human capital of children is stressed in the literature as an important
aspect of investment decisions of remittance receiving households. A sizeable number of
studies focuses on the impacts of migration and remittances on educational attainment of
children. Cox-Edwards and Ureta (2003) find that remittances reduce the school dropout hazard
rates of 6 to 24 years old boys and girls in El Salvador. Acosta (2011), also in the case of El
Salvador, finds on average null effect of receiving remittances on the likelihood of children
between ages 10 and 18 attending school. When the differences by demographic groups are
taken into account, girls between ages 10 and 18 from remittance receiving households are
around 10% more likely to attend school compared to non-recipient counterparts, yet the null
impact remains the same for boys between ages 10 and 18. Considering the differences by age
groups, remittance receiving children between ages 15 and 18 seem to be less likely to attend
school compared to non-recipient counterparts, whereas the evidence suggests no difference in
school attendance with respect to remittance-receipt status of the household for children of ages
3
10 to 14. The former and the latter studies use consecutive waves of the nationally
representative cross-sectional household survey for El Salvador (Encuesta de Hogares de
Propositos Multiples—EHPM) for years 1997 and 1998, respectively. In addition to the
differences in estimation samples, above studies also differ in that Cox-Edwards and Ureta
(2003) do not address endogeneity of remittances. Yang (2008), in the case of Philippines, states
that positive exchange rate shocks for international migrants lead to enhanced human capital
accumulation in origin households. His results support the claim that remittances increase child
school attendance and educational expenditure. He concludes that a positive exchange rate
shock for international migrants is associated with an increase in school attendance rates of 10
to 17 years old girls. However, there is no such causal relationship between positive exchange
rate shocks and school attendance rates of 10 to 17 years old boys. Bansak and Chezum (2009)
show that, in Nepal, remittances increase school attendance of young children (5 to 10 years
old males and females) with the effect being larger for males. They also show that receiving
remittances do not change the likelihood of school attendance of old children (11 to 16 years
old males and females). Lopez Cordova (2005), in the case of Mexico, provides evidence that
remittances decrease illiteracy rates of children aged 6-to-14, and increase school attendance
rates of five years old children. However, the impact on school attendance is insignificant for
6- to 14-years-old children and becomes negative for children between 15 and 17. Lopez
Cordova (2005) doesn’t investigate whether there is heterogeneity in the impacts of remittances
with respect to the gender of the child. Hanson and Woodruff (2003) tried to identify a causal
linkage between child schooling and having a household member living abroad for the case of
Mexico. Their results imply that 10- to 15-years-old girls whose mothers have less than 3 years
of schooling benefit from the migration of a household member with regard to accumulating
more years of schooling: additional 0.89 and 0.73 years of schooling for 10- to 12 and 13- to
15-years-old girls, respectively. They also show that migration has a positive impact on the
accumulated years of schooling for boys aged 10 to 12, but for this sample, the Sargan-Hansen
test rejects the null hypothesis that the excluded instruments are exogenous to the estimation
equation. Therefore, the results concerning boys aged 10 to 12 should be approached with
caution. Boys and girls aged 13 to 15 from migrant households where mothers have schooling
between 3 and 12 years obtain less schooling compared to their non-migrant counterparts. In
their study, years of schooling of the mother is used as a proxy for the wealth level of the
household. Hence, they argue that migration, via relaxing the household budget constraint
thanks to the remittances received, increase years of schooling attained for girls living in
households with low income levels. McKenzie and Rapoport (2011) investigate the overall
4
impact of migration on school attendance and the number of grade years completed for children aged 12 to 18 in rural Mexico. They find evidence of a negative significant effect of migration on school attendance and attainment. Their results show that living in a migrant household lowers the chances of boys completing junior high school and of boys and girls completing high school. Alcaraz, Chiquiar and Salcedo (2012), by exploiting the variation in the remittances—
due to 2008-2009 U.S. recession—that Mexican migrants’ origin households receive, find that a negative shock to the remittances is associated with a significant decrease in school attendance of 12 to 16 years old children left behind.
Outcomes related to child human capital accumulation is not restricted to child
schooling only. Child labor is an outcome as important as child schooling with respect to child
human capital investment decisions of households. Labor force participation of a child reduces
the time available to spend on education. There is a consensus in the literature regarding the
negative correlation between time spent on schooling and on labor for children. Hanson and
Woodruff (2003) argues that in poor countries, while deciding on the schooling of the child, the
main cost for the household is not the tuition, books, or uniforms but the foregone earnings of
the child. Households which would not rely on their children’s wage labor are those that can
maintain a satisfactory wealth level. In the light of these explanations, increasing educational
attainment of children may come through decreasing their participation in labor force and this
can be achieved by increasing the income level of households. As a priori guess, remittances
by increasing household budget and relaxing liquidity constraints of households may serve this
function. There is a large literature on how remittances affect child labor. Yang (2008) makes
use of an exogenous variation in origin household’s income which results from exchange rate
shocks to Filipino migrants and concludes that an increase in the size of the exchange rate shock
is associated with a decline in total hours worked by 10 to 17 years old males, while there is no
significant association between positive exchange rate shocks and total hours worked by 10 to
17 years old girls. When the composition of the work done is considered, boys aged 10 to 17
work fewer hours in unpaid family work, and work more hours in self-employment; however,
the increase in hours worked in self-employment is not enough to cover the overall decrease in
total hours worked for boys. An increase in the exchange rate shock is associated with a
decrease in hours worked in unpaid family work for girls aged 10 to 17 but the impact is only
marginally significant at 10% level. McKenzie and Rapoport (2011), in the case of Mexico,
investigate the reason of lower levels of school attendance and years of schooling accumulated
for migrant families’ children and find as an explanation doing housework for girls between
ages 16 and 18 and migrating themselves for boys at all age cohorts (12 to 15, and 16 to 18
5
years old). There isn’t a significant effect of having a migrant household member on 12 to 18 years old boys’ likelihood of working as unpaid family workers or as wage earners. Their study reveals that girls between ages 16 and 18 lose on both dimensions—schooling and work experience—of human capital accumulation. In other words, 16 to 18 years old girls from recipient households have lower rates of school attendance and less work experience compared to 16 to 18 years old girls from non-recipient households. Acosta (2011), in El Salvador, finds on average that remittances decrease the likelihood of working for wage and increase the likelihood of working as non-wage laborer (i.e., doing unpaid family work) for children aged between 10 and 18. When the differences in genders are accounted for, girls aged between 10 and 18 from recipient households seem to have lower chances of working for wage or working as non-wage laborer compared to their non-recipient counterparts. Boys aged between 10 and 18 from migrant households are substituting wage labor with non-wage labor. Alcaraz, Chiquiar and Salcedo (2012) find a significant increase in child labor resulting from a decrease in remittance receipts for 12 to 16 years old children in Mexican migrants’ origin households.
A substantial part of the literature on the economic impacts of remittances is concerned
with the linkage between remittances and labor force participation decisions of adults in origin
households. Theoretically, the direction of the impact of remittances on adult labor force
participation decisions is uncertain. If migrants’ earnings abroad are substantially higher than
their corresponding domestic labor market earnings potential, then the remittances sent by the
migrants may positively affect the household income. As any other source of non-labor income,
remittances will increase the reservation wages of the non-migrants in the household. This
income effect may direct non-migrants in the household to substitute labor with leisure: increase
the likelihood of leaving the labor force or give them enough motives to stay out of it at all
(Killingsworth, 1983). However, if the remittances are channeled to maintain existing
household enterprises or to set up a new household enterprise, then there may be an increased
demand for labor in the migrants’ origin households. This increased demand may reveal itself
in two ways: i) by an increase in the labor supply of non-migrants in the household either by
new entry to the labor market as non-wage (i.e., self-employed, employer, or unpaid family
worker) or wage laborer; or by an increase in the working hours of non-migrants who already
participate in the labor force; and ii) by substitution of non-wage labor for wage labor, or by a
shift from a category of non-wage work to another category of non-wage work (from unpaid
family work to self-employment or vice versa). Increased demand for the non-migrants’ labor
may deduce from the necessity of replacing the absent migrant’s labor and/or income as well
as from the productive use of remittances through financing household enterprises which
6
creates its own demand for additional manpower (Binzel and Assaad, 2011). Remittances, by creating opportunities to enter the labor force may give rise to earning own income, and as a result may benefit the non-migrant women in obtaining a higher bargaining power in the household. The empowerment of women in the household may have impacts on the child schooling and child labor decisions, may change the allocation of resources in favor of children;
thus, benefit the child human capital accumulation. It is, therefore, necessary to understand the impacts of remittances on the labor force participation decisions of adults and especially of women. Binzel and Assaad (2011), find a significant increase, resulting from the migration of a male household member, in the likelihood of participating in the labor force and in the likelihood of working as non-wage laborer (self-employed, employer or unpaid family worker) for women aged 25-49 in rural Egypt. In the intensive margin, they could not find a significant impact of migration on the working hours of the women left behind. Acosta (2006), in El Salvador, finds a significant negative impact of receiving remittances on the labor force participation of women in the migrants’ origin households: women from remittance receiving households are 60 percentage points less likely to participate in labor market. Lokshin and Glinskaya (2009) examines the impact of male migration on prime-age women’s labor force participation rates in Nepal and finds that migration of a male household member reduces labor force participation rates of women by 5.3 percentage points. Mendola and Carletto (2009) accounting for remittances effect, find a significant negative impact of current migration experience on the probability of engaging in paid self-employment (of size 54%) and a significant positive impact on the probability of engaging in unpaid work (of size 32%) for Albanian women. There is no impact of migration experience in the household on Albanian men’s labor force participation decisions, though. Amuedo-Dorantes and Pozo (2006) by accounting for the endogeneity of remittance income, show that Mexican men reduce work hours in formal sector (i.e., wage and salary work with a contract) and in urban self- employment, and increase work hours in informal sector (i.e., wage and salary work without a contract) due to an increase in the amount of remittances received. Thus, they argue that the disruptive effect of out-migration of a household member outweighs the income effect of remittances on labor supply behavior of males left behind—the forgone income or labor of the migrant, besides related migration costs, seems to be compensated by an increase in the labor supply of other male members of the household in informal sector. Increase in remittances is associated with a decrease both in unpaid family work and in informal work for Mexican women, suggesting dominance of income effect over disruptive effect of remittances.
Rodriguez and Tiongson (2001), in Philippines, find that migration reduces labor market
7
participation of 15- to 64-year-old males and females. Nevertheless, their definition of labor market participation includes paid employment and self-employment but excludes unpaid family work, plus the endogeneity of migration is not addressed in their analysis. Cox-Edwards and Rodriguez-Oreggia (2009), in Mexico, find that receiving persistent remittances do not affect labor supply behavior of either men or women of ages 12- to 65-year-old. They argue that the migrant sends back remittances to recover his pre-migration contribution to the household income and the amount of remittances sent is not large enough to alter the prices of labor to achieve a significant difference in non-migrants’ labor supply behavior between receiving and non-receiving households.
While a large fraction of the literature on the impacts of remittances is dedicated to human capital accumulation outcomes—child or adult—some focus on the impacts on household well-being. Adams (1998), in the case of rural Pakistan, is unable to find any significant impact of remittances on non-farm asset accumulations. Lopez Cordova (2005) shows that, in Mexico, receiving remittances decreases the chances of households suffering from poverty where poverty is defined as the household income being at most two times of the official minimum wage. However, remittances do not have a significant impact on extreme poverty where extreme poverty cutoff is set at the official minimum wage. Lopez Cordova (2005) argues that high costs associated with international migration is the main reason behind the finding of a zero impact of remittances on extreme poverty. Households suffering from high levels of poverty cannot afford to migrate and send remittances back home. His findings suggest that there is a lower boundary of income for a household to benefit from migration and remittances. Adams and Page (2003), on the other hand, analyze seventy-four countries and show that a 10 percent increase in the amount of remittances received decreases the share of people living under 1 dollar per day by 1.9 and 1.6 percentage points in low and middle-income countries, respectively.
The rest of the paper is organized as follows: the next subsection provides information
on migration history of Turkey with a special focus on the important role that remittances play
on economic development. Section 1.2 presents the identification strategy, the empirical
approach, and a thorough examination of the issues accompanying estimation of impacts of
remittances on binary household outcomes. Section 1.3 describes the data and presents
descriptive statistics before carrying on with the estimation results in Section 1.4. Finally,
Section 1.5 concludes.
8
1.1.1 Migration and remittance history of Turkey
In the beginning of 1960s, Turkey was experiencing an unemployment rate of 10 percent and an additional underemployment over 15 percent (Icduygu, 2009). Turkish government borrowed heavily from other countries and had difficulties in paying its debts due to the foreign currency bottlenecks (Icduygu, 2009). At the same time, industrialized European countries were in serious need of manpower. In light of these developments, Turkey signed bilateral agreement with Federal Republic of Germany in 1961 that allowed emigration of workers from Turkey to Germany (Koc and Onan, 2004). This was the leading step for the mass emigration of Turkish workers to European countries. The main motivations for the Turkish government in promoting emigration were to reduce unemployment and gain foreign currency through remittances (Icduygu, 2009).
With the opening of the corridor of emigration in 1961, the number of workers going to Europe increased dramatically and peaked at 66,000 people in 1964 (Icduygu, 2009). Till the oil crisis of 1974, mass emigration to Europe continued. 1975 is the last year of observed mass emigration to Europe (Icduygu, 2009). The European countries were deeply affected from the oil crisis and they stopped accepting immigrant workers. Turkish government, then, tried to find new destination routes for its excess supply of labor. The new destination was set to be oil rich Arab countries. Immigrant workers in Arab countries were hired for a specified amount of time—till their assigned project ends—and they were not allowed to bring their families with them (Icduygu, 2005). Over the period of 1975-1980, more than 75,000 contracted workers had gone to the oil-exporting countries (Icduygu, 2009). However, by the mid-1990s, due to the completion of large-scale infrastructural projects most of the immigrant workers had to turn back to Turkey.
With the collapse of USSR in the 1990s, newly emerging countries started reconstruction programs and demanded labor. The mid-1990s experienced mass emigration to CIS countries (former Soviet Republic countries) with a total of 65,000 emigrants (Icduygu, 2009).
In the early 2000s, while Turkey’s population was around 70 million, the emigrants had
a total of about 3.5 million. The largest share of emigrants was residing in Europe, a total of 3
million, followed by 300,000 emigrants in Australia, Canada and U.S. The next largest emigrant
receiving region is CIS countries with a total of 150,000. Lastly, around 100,000 emigrants
were present in Arab countries (Icduygu, 2005). International migrants constituted 5 percent of
Turkey’s population.
9
Between 30 to 40 percent of past emigrants permanently returned back to Turkey (Icduygu, 2005). Besides having 5 percent of the population as current emigrants, this implies that a nonnegligible portion of the population in Turkey has direct migration experience. In addition, emigrants don’t lose their contacts with the families left behind and many of them send remittances. A huge migration experience of this sort could potentially have some effects on home country’s economy and migrants’ origin households.
The most striking impact of emigration on Turkey’s economy is through remittances.
From 1960s to 2000s, accumulated value of remittances is $75 billion. In 1967, remittances amounted $93 million. In 1974, the corresponding figure was $1.4 billion and, in 1978 remittances amounted $893 million. Between 1978 and 1988 average annual remittances amounted to 1.5-2 billion dollars. In 1980s, remittances amounted 65 percent of trade deficit and 2.5 percent of GNP. During late 1980s and early 1990s, average annual remittance receipt was about $3 billion with a peak of $3.4 billion in 1995 (Icduygu, 2009). In 1990s remittances amounted one third of the trade deficit and less than 2 percent of GNP. In late 2000s, remittances help to cover only around 2% of Turkey’s trade deficit. Obviously, it cannot be suggested that the decrease of remittance share of trade deficit and GNP is due to the decrease in annual remittance amounts. The decrease in the share of trade deficit and GNP can be explained with the growth of Turkish economy and lower contribution of remittances in the corresponding shares compared to the contributions from tourism, exporting and other income sources (Icduygu, 2005). It is an undeniable fact that remittances played a major role in financing the import bill of Turkey since 1960s. In addition to providing foreign currency through remittances, emigration also relieved the pressure on unemployment rates. Turkey had experienced an unemployment rate of 16.7% in 1986 and it is argued that the unemployment rate would have reached 23.2% in 1986 in the absence of labor emigration (Barisik et al., 1990).
As a result, it can be argued that a successful policy was run in Turkey to overcome the foreign currency bottlenecks and to reduce unemployment.
Even though Turkey has an impressive migration history and has accumulated significant amounts of remittances, there are very few studies regarding the impacts of international migration and remittances in the context of Turkey.
There is a well-known migration study in Turkey; 1996 Turkish International Migration
Survey (TIMS-96). Data was collected from 28 selected districts in 8 provinces of Turkey in
1996 and was not representative at the national level. According to TIMS-96, 12 percent of
households received remittances and 80 percent of remittance receiving households used
remittances to improve their standard of living. In TIMS-96, there is also evidence for regional
10
differences in the amount of remittances received. It is found that households located in less developed regions are more likely to receive remittances than households in developed regions.
Koc and Onan (2004), by using data from TIMS-96, find that remittances are basically used to satisfy consumption needs of origin households. This is a conflicting result with findings of Yang (2008) who shows that increased remittance income deriving from international migrants’
exchange rate shocks is not associated with any change in consumption of origin households in Philippines. Koc and Onan (2004) also show that remittance receiving households are better off than non-recipient households. This implies that remittances have a positive impact on household welfare. Day and Icduygu (1999) use data gathered from 234 individuals in Turkey during 1992-1993 and show that return migrants and their close relatives have higher consumption levels than non-migrants. Keles (1985) conclude that remittances do not work in the direction of reducing imbalances between regions of Turkey, but benefit the remittance receiving households via improving their standards of living. Atalik and Beeley (1993) find that remittances are used for investment in physical capital such as acquisition of land, and cars.
In the Turkish context, there are some studies on the determinants of remittances
1, to the best of our knowledge, impacts of remittances on different aspects of human capital accumulation were not studied thoroughly. In addition, datasets that were used in prior studies on impacts of migration and remittances in the context of Turkey were not nationally representative which poses problems for the external validity of the estimates. This study, by implementing a nationally representative micro level dataset, aims to fill this gap and contribute to the literature by studying the impacts of remittances on child schooling, child illiteracy, child labor, adult labor force participation and household wellbeing in the context of Turkey.
Furthermore, much attention has been paid to solve econometric problems associated with estimation of binary choice models with a binary endogenous variable.
1.2 Identification Strategy and Estimation Methodologies 1.2.1 Econometric Identification
Hoddinott (1994) states that migration decision is an outcome of a utility maximization problem of the household solved jointly by the prospective migrant and the other household members. Thus, the allocation of migrants and migrant earnings across households may not be random. The main empirical challenge in consistently estimating the causal impacts of remittances on schooling/labor outcomes is due to a possible correlation of households’
1 Aydas et al. (2005), Köksal (2006), Van Dalen et al. (2005)
11
remitting behavior with unobserved determinants of outcomes. For instance, the education mobility literature presents evidence of a positive association between parents’ heritable schooling endowments and their children’s educational attainment
2. In addition to this linkage between parents’ ability and their children’s schooling outcomes, if parents with higher heritable genetic ability find it more appealing to migrate and remit in order to finance their children’s schooling expenses, then a simple comparison of remittance receiving and non- receiving families overestimates the impacts of remittances on child schooling. Hanson and Woodruff (2003) presents a different scenario: households which experience negative income shocks may decide to send a member abroad to cover the financial losses. Children in such households may need to reallocate their time favoring labor over schooling to compensate for the short-term income shortages resulting mainly from the unfavorable shock. A comparison of remittance receiving and non-receiving families, in that case, will understate the education gains from remitting. Therefore, the direction of endogeneity bias is uncertain. Moreover, reverse causality problem may arise if families consider migration and remittances as a leeway for funding their children’s education.
To solve the endogeneity problem of remittances, we implement an instrumental variable estimation strategy and follow McKenzie and Rapoport (2011) and a number of studies
3in using regional level historical migration rates as instruments. International migration incurs some substantial costs: monetary costs related with transportation, costs of acquiring information about the destination country, opportunity costs in terms of lost income while searching for jobs in the destination country, and psychic costs related with losing contact with parents, beloved ones, friends and relatives (Massey, 1988). Migration networks help lowering these costs by providing a prospective migrant information and help about ways to enter the destination country, finding job, accommodation and adapting to a new culture. Households with better access to migrant networks bear a lower cost of migration, and thus, are more likely to migrate and send remittances. Migration stocks become self-perpetuating via cost lowering impact of migrant networks, and thus, migration networks formed earlier affect migration decisions of households today (Munshi, 2003; McKenzie and Rapoport, 2007). Migration rates are argued to be good indicators of migration networks present in a village, municipality or a
2 Holmlund, Lindahl and Plug (2011), Behrman and Rosenzweig (2002)
3 Hanson and Woodruff (2003), McKenzie and Rapoport (2011), Hildebrandt and McKenzie (2005) all use historical migration rates as instruments to predict current migration stocks. Acosta (2011) uses historical migration rates as instruments for receiving remittances on the household side. Alcaraz et al. (2012) and Lopez Cordova (2005) use the placement of rail lines in Mexico in 1920 -the distance from municipality to the rail road plus the distance from the rail road to the US-Mexico border-, which mainly captures migration networks present in municipalities, to instrument current remittance receipts.
12
state (Hanson and Woodruff, 2003). Consequently, historical migration rates may serve as instrumental variable for current migration decision of households and remittance receipts. The migration rate we use comes from the 1985 Turkey Census data and is calculated as the share of international migrants in a region’s population. The international migrants are defined as those Turkish citizens who had changed their residence country to Turkey from a host country during the previous five years. There are 26 regions in Turkey which are statistical agglomerates of provinces and each region consists of provinces which are similar in characteristics such as population, socioeconomic development, geography, per capita GDP, per capita output in industry, agricultural output, and urbanization rate
4. We calculate historical regional migration rates by taking a weighted average of 1985 migration rates of provinces in a given region where a province’s weight is equivalent to the population share of that province in the region.
IV estimation relies on mainly two assumptions; migration networks should be strongly correlated with remittances, and migration networks should not affect potential outcomes other than their impact through remittances. These assumptions are known as existence of first stage and exclusion restriction of the instrument, respectively. Existence of first stage is argued to hold via cost lowering impacts of migration networks which induce continuing waves of migratory movements from a region and continuing remittance receipts in a region because of the sustained migration. Furthermore, existence of first stage could be confirmed through running a regression of remittance receipts on migration networks. Many studies successfully establish a strong correlation between migration networks and remittance receipts
5. The challenging part is to justify the exclusion restriction of the instrument.
There are some potential threats to the exclusion restriction; hence, to the validity of the instrument. Woodruff and Zenteno (2001) is one of the pioneers in forming the sociological linkage between migration networks and current migration flows. Studies thereafter make use of the instrument in predicting current migration and remittances. However, for migration networks to instrument remittances, one needs to assume that the only impact of migration networks on outcomes is through remittances. If there are other impacts of migration on outcomes that are distinct from remittances, then the error term will capture these impacts and migration network will eventually be correlated with the error term since migration networks are good predictors of migration (McKenzie, 2005). The literature presents evidence on migration having impacts different than and most likely conflicting with its most apparent
4 The regional classification used in the study is provided by TÜİK and is at NUTS-2 level.
5 Acosta (2011), Alcaraz et al. (2012), Lopez Cordova (2005), Bansak and Chezum (2009), Cattaneo (2012)
13
impact; remittances. Hanson and Woodruff (2003) note that migration may disrupt the family structure and leave children in migrant households without a guardian or a role model. In addition, children may be forced to participate in labor market to compensate for the lost income of the migrant family member. McKenzie and Rapoport (2011) state that children of migrant families are more likely to migrate than children of non-migrant families. If there are differences in returns to education in source and host countries, this may incentivize children of migrant families to substitute education with migration or vice versa. Trying to isolate the impact of remittances from other impacts of migration by means of instrumenting remittances with migration networks may lead the instrument to capture these other impacts of migration which is a violation of exclusion restriction (McKenzie, 2005). Thus, to estimate pure impact of remittances (income effect), one may need to account for other impacts of migration while instrumenting remittances with migration networks
6, or try to find an instrument which predicts not only why one household is more likely to have a migrant member compared to an observationally similar household, but also why one migrant family sends more remittances compared to an observationally similar migrant family
7. Hence, most of the studies that instrument remittances with migration networks are likely to estimate the combined impact of remittances and other impacts of migration
8. In other words, they implicitly estimate the overall impact of migration. In this study, we acknowledge that the main interest is not to estimate pure monetary impacts of remittances, instead it is argued that the remittance receipt status is a good proxy for the migration experience of a household. In our data, among remittance receiving households 36.5% of them have either a missing male or a missing female spouse of the household head. Although, information about the migration experience of household members or information about the household members who are absent at the survey date are not available, we are inclined to think that the missing male or female spouse is the source of the remittances.
This assumption, if true, may imply either a recent or a past migration experience for a household. Regarding this last point, 11% of remittance recipient households with a missing male spouse, receive remittances in the form of pension benefits only, and almost 90% of
6 Bansak and Chezum (2009), in the case of Nepal, account for the household disruption impact of migration by controlling for the number of adults living outside the household.
7 Yang (2008), in the case of Philippines, induces exogenous variation in amount of remittances through exchange rate shocks which are argued to be randomly distributed over migrant households. Yang (2008) considers only the households with international migrants before the unexpected Asian financial crysis in 1997 and uses the change in the exhange rate as the treatment which is argued to be randomly distributed across migrants and shows that the elasticity of remittances with respect to exchange rate is 0.6.
8 Bansak and Chezum (2009), Yang (2008) and Lopez Cordova (2005) differ from the rest of the studies as their identification strategies try to isolate the impacts of remittances from other consequences of migration.
14
female household heads in these households have either lost or divorced their spouses. The corresponding share of households with missing female spouses which receive remittances in form of pension benefits only is 37%, and 74% of the male household heads either have lost or divorced their spouses. These statistics lead us to imagine that either the spouses that we observe in the data or their counterparts had a past migration experience. For households with a missing male spouse which receive remittances in forms other than retirement pensions, 67% of the left behind female partners are married and 23% of them have a passed away spouse. This may suggest that the husbands or the husbands’ close relatives may provide the female household heads with remittances, even though there is no evidence to prove the latter. This observation, contrary to the preceding case, suggests a recent migration impact on the household and provides us with more confidence in assuming that the source of the remittance is the spouse living outside the house. Another channel to link remittances with the migration experience of household is through the observation that recipient households of pension benefits from a host country where both partners are at home constitute 27% of all remittance receiving households.
This is an indicator for return migration within the household. In total, almost 64% of remittance receiving households can be linked to the source of the remittance, and hence, can be attributed with a possible migration experience. The rest of the recipient households may receive the remittances from other household members (e.g. children, or grandchildren), close relatives to the family, or friends albeit there is no way to confirm the source of the remittances in this case.
Acosta (2011) uses cross sectional data from El Salvador and finds that nearly 30% of recipient households receive remittances from outside their circle of close relatives. Our data is in line with Acosta (2011) regarding the relationship of the remitter with the family left behind. Still, the descriptive evidence suggests a strong relation between remittances and households’
migration experience.
Migration is defined in a variety of ways in the literature. Some use narrow definitions of migration
9which is good at unraveling the impacts of migration that may have occurred concurrently with the incidence of migration. Some use broader definitions of migration
10. Migration may have long-lasting impacts, that is, the impacts may have been preserved for a long period of time. The change in household resources due to the migration of a household member six or more years ago may still affect the households’ schooling decisions for their children. The negative impacts of a household member’s migration six or more years ago, for
9 Hanson and Woodruff (2003), Lopez Cordova (2005) define migration as the change of residency within last 5 years from source country to a host country.
10 McKenzie and Rapoport (2011) define migration as ever been to another country for work or other reasons.
15
example, might have forced children to leave the school when they were young and it is likely that these children will not be observed at school when they get older.
This scenario bears an impact of migration at the extensive margin, and migration defined as change of residency within the last 5 years or having a current migrant member may be insufficient to reveal the impacts of migration at the extensive margin. Especially, when the interest is to find the impacts of migration on schooling and child labor, making use of the broad definition of ever migrating allows to estimate such persistent impacts. The remittance variable in our study, by the inclusion of households that receive remittances as retirement pensions besides in cash and in-kind, capture both past, recent and current migration experience of households, in other words, whether households have ever engaged in migration. To sum up, remittance receipt of households is argued to capture not only income effect of remittances but also other consequences of migration experience, and is used to estimate average impact of ever migrating on households. To be precise, IV methods estimate local average treatment effects;
the effects on the group of compliers. These are the units that take the treatment when they are exposed to the instrument, and do not take the treatment when they are not exposed to the instrument. In our study, households that receive remittances when they have a large migrant network and do not receive remittances when they have a small migrant network comprise the complier group. McKenzie and Rapoport (2007) show that these households come from the lower end of the wealth distribution as they cannot afford migration unless they have access to a large migration network that help reduce the migration costs hugely. This is a group worth investigating the impacts of migration because remittances may benefit them more compared to households that come from upper parts of the wealth distribution.
Another threat for the validity of the instrument is that migration networks measured by regional migration rates in 1985 can predict outcomes through means other than remittances.
This is possible, in particular, if there were regional characteristics that influence migration
historically and persist to influence outcomes of interest today. Being unable to account for all
possible channels distinct from remittances, through which migration networks may explain
part of the variation in outcomes, results in violation of the exclusion restriction and renders the
instrument invalid. The initial emigration to Europe from Turkey between 1961 and 1974,
which helped creating migration routes that prospective migrants follow, was heavily organized
by Turkish Employment Service (TES). Unless a specific worker is demanded by the employer
in the host country, an individual had to apply to a local TES office located mostly at city centers
and register his name in a waiting list. Whenever a job position opens, it was offered to the
relevant candidate in the waiting list by TES (İçduygu, 1991). Back then, the Turkish
16
government desired mass flows of emigration to Europe in order to reduce unemployment, gain foreign exchange earnings through remittances, and provide grounds for development projects for the underdeveloped regions of the country. With regard to the last point, emigrants from relatively poor regions of the country and from regions of natural disasters were prioritized by TES to migrate at once (Abadan-Unat, 2006). Day and İçduygu (1997) comment on the relationship between socioeconomic development of regions and emigration in Turkey, and show that when the regions get poorer emigration increases, but when socioeconomic development falls below a certain level, emigration levels start to decline as well. It can be considered within this context that in 1960s and 1970s, some poorest cities in less developed Eastern Anatolia region never achieved to become significant emigration sources whereas, relatively poor cities like Denizli, Afyon and Yozgat from more developed Western and Central Anatolia regions were the main sources of emigration to Europe (Ayhan et al. 2000).
Apparently, the creation of migrant networks was influenced partly by regional disparities in development levels, and it is implausible to assume that migration networks are distributed randomly across regions. If historical inequality in development levels, besides helping determine the migration networks, also continue to influence schooling and labor today then it is necessary to account for historical levels of inequality to preserve the validity of the instrument. Historical schooling differences between regions may also pose problems for the exogeneity of the instrument. If historical schooling levels vary accordingly with migrant networks and have impacts on current schooling levels (i.e., through intergenerational transmission of schooling), then not addressing this channel would invalidate the instrument. It is likely to observe a relation between historical schooling levels and historical migration networks as schooling levels are good predictors for level of development in a region, and it is shown that historically high emigration regions are less developed (Ayhan et al. 2000). To account for historical schooling and historical inequality levels, we control for regional measures from around the same time period as our instrument: length of road per 1 km
2by region in 1980, the share of length of asphalt roads in total length of roads by region in 1985
11, the interaction between these two variables, development index values from 1973
12, school
11 The information necessary to create road related variables is acquired from General Directorate of Highways Maintenance Division.
12 DİE (Devlet İstatistik Enstitüsü) 1973: 72-5. DİE estimates the development index for each province considering the following indicators: proportion of urban population, literacy rate, number of high school and university graduates, paid income tax per capita, number of hospital beds per 100,000 persons, number of persons per radio, length of road per 1 km2, average number of workers per workplace, per capita added value, per capita industrial added value, proportion of agricultural workers in total workforce, and share of industrial laborers in total workforce. We take a weighted average of provinces’ index values with respect to their population shares in the region (1970 Census is used to gather the information on provinces’ populations).
The higher the index value the higher the development level is. The index value for the country is standardized at 1.
17
attendance rates for males and females aged between 6 and 10 by region in 1985, number of schools per 1,000 children aged between 6 and 16 by region in 1985
13.
Turkish government aimed to increase the standard of living for citizens residing in relatively poor regions or natural disaster areas by giving them priority in migration and facilitating their move; however, the only interest of the government was not the welfare of the emigrant households. Turkish government also considered migration and remittances as a means of speeding up the development process of underdeveloped regions, and in order to achieve this goal tried to channel migrants’ earnings to employment generating activities in less developed regions via the installation of three development programs in 1970s (Keleş, 1985;
Martin, 1991). Firstly, Worker’s Joint Stock Companies were founded to foster the development in less developed regions and hence, reduce regional disparities. Migrants’ remittances and non- migrant households’ contribution in migrant sending regions were the main two sources of financing these institutions. The investments made by these institutions would benefit returning migrants in finding jobs and serve as a tool to develop regions of origin. Secondly, Turkish government initiated the establishment of Village Development Cooperatives in mostly poor regions of the country, and remittances served as one of the main funding through which Village Development Cooperatives operate. These Cooperatives had a nonnegligible impact on the development of various migrant sending regions; as an example, Boğazlıyan, which is a town of Yozgat and is one of the main migrant sending regions to Europe, experienced a rapid increase in the number of agricultural machineries from 300 in 1966 to 1,500 in 1975 thanks to investments made possible by migrants’ earnings (Abadan-Unat et al. 1976). Besides improving backward regions, these Cooperatives also offered migration possibilities to its members, since members of Village Development Cooperatives in poor regions had priority in migrating, and the huge increase in the number of Cooperatives from 2,000 in 1971 to almost 6,000 in 1974 reveals the role of Cooperatives in easing migration (Abadan-Unat, 2006). It can be argued that there is a two-way relation between migration from and development of migrant sending regions: more migration might have accelerated development of the source regions for migrants through investments of government initiated development programs which as discussed mainly operate on migrant earnings, and these institutions besides contributing to the development process of emigrants’ source regions might have increased the stock of migrants from these regions via facilitating migration. Lastly, State Industry and Workers’ Investment Bank was
13 We benefit from National Education Statistics 1985-1986 which is published by DİE with regard to number of students enrolled and number of schools for related age categories, and for the total number of children in given age categories we make use of 1985 Turkey Census.
